Research Area:
Simulation Methods, Macroeconometrics, Inference in Econometrics
Tests for structural change (Andrews (1993); Andrews, Lee and Ploberger (1996)) have relied on asymptotic critical values that depend on the type of model as well as on trimming rules over which the tests are computed. In practice, these dependencies may be inconvenient. We investigate the sensitivity of tests for structural change to the choice of trimming rule using bootstrap methods. In terms of size, we find that the the bootstrap tests, in comparison to the asymptotic tests, have empirical sizes closer to their nominal sizes and that trimming has no effect. Bootstrap tests for structural change have low power for all trimming rules but larger power than their asymptotic counterparts. Power is highest when the break is within the trimming rule, and exponential tests have always greatest power. Lowest power occurs when the break is outside the trimming rule and when the break is at the end-points. In practice one should use the bootstrap exponential tests and a narrow trimming rule if the break is believed to be near the middle of the sample and a wider one if it is not. We illustrate our findings with an empirical example on the growth rate of Canadian and US GDP.
This paper investigates the behaviour of estimators based on the Kullback-Leibler information criterion (KLIC), as an alternative to the generalized method of moments (GMM). We first study the estimators in a simulation model of consumption growth with power utility. Then we compare KLIC and GMM estimators in macroeconomic applications, in which preference parameters are estimated with aggregate data. KLIC probability measures serve as useful diagnostics, but, in dependent data, tests of overidentifying restrictions in the KLIC framework do not generally have size properties superior to those of iterative GMM.
e-mail: lamarchj@qed.econ.queensu.ca